?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'cleancache/linux-next'
[linux-2.6/next.git] / drivers / net / iseries_veth.c
blob9ece1fd9889dbd1d40f427c8831fb1433f5d1e9c
1 /* File veth.c created by Kyle A. Lucke on Mon Aug 7 2000. */
2 /*
3 * IBM eServer iSeries Virtual Ethernet Device Driver
4 * Copyright (C) 2001 Kyle A. Lucke (klucke@us.ibm.com), IBM Corp.
5 * Substantially cleaned up by:
6 * Copyright (C) 2003 David Gibson <dwg@au1.ibm.com>, IBM Corporation.
7 * Copyright (C) 2004-2005 Michael Ellerman, IBM Corporation.
8 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License as
11 * published by the Free Software Foundation; either version 2 of the
12 * License, or (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
22 * USA
23 *
24 *
25 * This module implements the virtual ethernet device for iSeries LPAR
26 * Linux. It uses hypervisor message passing to implement an
27 * ethernet-like network device communicating between partitions on
28 * the iSeries.
29 *
30 * The iSeries LPAR hypervisor currently allows for up to 16 different
31 * virtual ethernets. These are all dynamically configurable on
32 * OS/400 partitions, but dynamic configuration is not supported under
33 * Linux yet. An ethXX network device will be created for each
34 * virtual ethernet this partition is connected to.
35 *
36 * - This driver is responsible for routing packets to and from other
37 * partitions. The MAC addresses used by the virtual ethernets
38 * contains meaning and must not be modified.
39 *
40 * - Having 2 virtual ethernets to the same remote partition DOES NOT
41 * double the available bandwidth. The 2 devices will share the
42 * available hypervisor bandwidth.
43 *
44 * - If you send a packet to your own mac address, it will just be
45 * dropped, you won't get it on the receive side.
46 *
47 * - Multicast is implemented by sending the frame frame to every
48 * other partition. It is the responsibility of the receiving
49 * partition to filter the addresses desired.
50 *
51 * Tunable parameters:
52 *
53 * VETH_NUMBUFFERS: This compile time option defaults to 120. It
54 * controls how much memory Linux will allocate per remote partition
55 * it is communicating with. It can be thought of as the maximum
56 * number of packets outstanding to a remote partition at a time.
57 */
58
59 #include <linux/module.h>
60 #include <linux/types.h>
61 #include <linux/errno.h>
62 #include <linux/ioport.h>
63 #include <linux/kernel.h>
64 #include <linux/netdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <linux/init.h>
68 #include <linux/delay.h>
69 #include <linux/mm.h>
70 #include <linux/ethtool.h>
71 #include <linux/if_ether.h>
72 #include <linux/slab.h>
73
74 #include <asm/abs_addr.h>
75 #include <asm/iseries/mf.h>
76 #include <asm/uaccess.h>
77 #include <asm/firmware.h>
78 #include <asm/iseries/hv_lp_config.h>
79 #include <asm/iseries/hv_types.h>
80 #include <asm/iseries/hv_lp_event.h>
81 #include <asm/iommu.h>
82 #include <asm/vio.h>
83
84 #undef DEBUG
85
86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88 MODULE_LICENSE("GPL");
89
90 #define VETH_EVENT_CAP (0)
91 #define VETH_EVENT_FRAMES (1)
92 #define VETH_EVENT_MONITOR (2)
93 #define VETH_EVENT_FRAMES_ACK (3)
94
95 #define VETH_MAX_ACKS_PER_MSG (20)
96 #define VETH_MAX_FRAMES_PER_MSG (6)
97
98 struct veth_frames_data {
99 u32 addr[VETH_MAX_FRAMES_PER_MSG];
100 u16 len[VETH_MAX_FRAMES_PER_MSG];
101 u32 eofmask;
102 };
103 #define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
104
105 struct veth_frames_ack_data {
106 u16 token[VETH_MAX_ACKS_PER_MSG];
107 };
108
109 struct veth_cap_data {
110 u8 caps_version;
111 u8 rsvd1;
112 u16 num_buffers;
113 u16 ack_threshold;
114 u16 rsvd2;
115 u32 ack_timeout;
116 u32 rsvd3;
117 u64 rsvd4[3];
118 };
119
120 struct veth_lpevent {
121 struct HvLpEvent base_event;
122 union {
123 struct veth_cap_data caps_data;
124 struct veth_frames_data frames_data;
125 struct veth_frames_ack_data frames_ack_data;
126 } u;
127
128 };
129
130 #define DRV_NAME "iseries_veth"
131 #define DRV_VERSION "2.0"
132
133 #define VETH_NUMBUFFERS (120)
134 #define VETH_ACKTIMEOUT (1000000) /* microseconds */
135 #define VETH_MAX_MCAST (12)
136
137 #define VETH_MAX_MTU (9000)
138
139 #if VETH_NUMBUFFERS < 10
140 #define ACK_THRESHOLD (1)
141 #elif VETH_NUMBUFFERS < 20
142 #define ACK_THRESHOLD (4)
143 #elif VETH_NUMBUFFERS < 40
144 #define ACK_THRESHOLD (10)
145 #else
146 #define ACK_THRESHOLD (20)
147 #endif
148
149 #define VETH_STATE_SHUTDOWN (0x0001)
150 #define VETH_STATE_OPEN (0x0002)
151 #define VETH_STATE_RESET (0x0004)
152 #define VETH_STATE_SENTMON (0x0008)
153 #define VETH_STATE_SENTCAPS (0x0010)
154 #define VETH_STATE_GOTCAPACK (0x0020)
155 #define VETH_STATE_GOTCAPS (0x0040)
156 #define VETH_STATE_SENTCAPACK (0x0080)
157 #define VETH_STATE_READY (0x0100)
158
159 struct veth_msg {
160 struct veth_msg *next;
161 struct veth_frames_data data;
162 int token;
163 int in_use;
164 struct sk_buff *skb;
165 struct device *dev;
166 };
167
168 struct veth_lpar_connection {
169 HvLpIndex remote_lp;
170 struct delayed_work statemachine_wq;
171 struct veth_msg *msgs;
172 int num_events;
173 struct veth_cap_data local_caps;
174
175 struct kobject kobject;
176 struct timer_list ack_timer;
177
178 struct timer_list reset_timer;
179 unsigned int reset_timeout;
180 unsigned long last_contact;
181 int outstanding_tx;
182
183 spinlock_t lock;
184 unsigned long state;
185 HvLpInstanceId src_inst;
186 HvLpInstanceId dst_inst;
187 struct veth_lpevent cap_event, cap_ack_event;
188 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
189 u32 num_pending_acks;
190
191 int num_ack_events;
192 struct veth_cap_data remote_caps;
193 u32 ack_timeout;
194
195 struct veth_msg *msg_stack_head;
196 };
197
198 struct veth_port {
199 struct device *dev;
200 u64 mac_addr;
201 HvLpIndexMap lpar_map;
202
203 /* queue_lock protects the stopped_map and dev's queue. */
204 spinlock_t queue_lock;
205 HvLpIndexMap stopped_map;
206
207 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
208 rwlock_t mcast_gate;
209 int promiscuous;
210 int num_mcast;
211 u64 mcast_addr[VETH_MAX_MCAST];
212
213 struct kobject kobject;
214 };
215
216 static HvLpIndex this_lp;
217 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
218 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
219
220 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
221 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
222 static void veth_wake_queues(struct veth_lpar_connection *cnx);
223 static void veth_stop_queues(struct veth_lpar_connection *cnx);
224 static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
225 static void veth_release_connection(struct kobject *kobject);
226 static void veth_timed_ack(unsigned long ptr);
227 static void veth_timed_reset(unsigned long ptr);
228
229 /*
230 * Utility functions
231 */
232
233 #define veth_info(fmt, args...) \
234 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
235
236 #define veth_error(fmt, args...) \
237 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
238
239 #ifdef DEBUG
240 #define veth_debug(fmt, args...) \
241 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
242 #else
243 #define veth_debug(fmt, args...) do {} while (0)
244 #endif
245
246 /* You must hold the connection's lock when you call this function. */
247 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
248 struct veth_msg *msg)
249 {
250 msg->next = cnx->msg_stack_head;
251 cnx->msg_stack_head = msg;
252 }
253
254 /* You must hold the connection's lock when you call this function. */
255 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
256 {
257 struct veth_msg *msg;
258
259 msg = cnx->msg_stack_head;
260 if (msg)
261 cnx->msg_stack_head = cnx->msg_stack_head->next;
262
263 return msg;
264 }
265
266 /* You must hold the connection's lock when you call this function. */
267 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
268 {
269 return cnx->msg_stack_head == NULL;
270 }
271
272 static inline HvLpEvent_Rc
273 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
274 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
275 u64 token,
276 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
277 {
278 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
279 HvLpEvent_Type_VirtualLan,
280 subtype, ackind, acktype,
281 cnx->src_inst,
282 cnx->dst_inst,
283 token, data1, data2, data3,
284 data4, data5);
285 }
286
287 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
288 u16 subtype, u64 token, void *data)
289 {
290 u64 *p = (u64 *) data;
291
292 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
293 HvLpEvent_AckType_ImmediateAck,
294 token, p[0], p[1], p[2], p[3], p[4]);
295 }
296
297 struct veth_allocation {
298 struct completion c;
299 int num;
300 };
301
302 static void veth_complete_allocation(void *parm, int number)
303 {
304 struct veth_allocation *vc = (struct veth_allocation *)parm;
305
306 vc->num = number;
307 complete(&vc->c);
308 }
309
310 static int veth_allocate_events(HvLpIndex rlp, int number)
311 {
312 struct veth_allocation vc =
313 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
314
315 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
316 sizeof(struct veth_lpevent), number,
317 &veth_complete_allocation, &vc);
318 wait_for_completion(&vc.c);
319
320 return vc.num;
321 }
322
323 /*
324 * sysfs support
325 */
326
327 struct veth_cnx_attribute {
328 struct attribute attr;
329 ssize_t (*show)(struct veth_lpar_connection *, char *buf);
330 ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
331 };
332
333 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
334 struct attribute *attr, char *buf)
335 {
336 struct veth_cnx_attribute *cnx_attr;
337 struct veth_lpar_connection *cnx;
338
339 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
340 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
341
342 if (!cnx_attr->show)
343 return -EIO;
344
345 return cnx_attr->show(cnx, buf);
346 }
347
348 #define CUSTOM_CNX_ATTR(_name, _format, _expression) \
349 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
350 { \
351 return sprintf(buf, _format, _expression); \
352 } \
353 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
354
355 #define SIMPLE_CNX_ATTR(_name) \
356 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
357
358 SIMPLE_CNX_ATTR(outstanding_tx);
359 SIMPLE_CNX_ATTR(remote_lp);
360 SIMPLE_CNX_ATTR(num_events);
361 SIMPLE_CNX_ATTR(src_inst);
362 SIMPLE_CNX_ATTR(dst_inst);
363 SIMPLE_CNX_ATTR(num_pending_acks);
364 SIMPLE_CNX_ATTR(num_ack_events);
365 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
366 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
367 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
368 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
369 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
370
371 #define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
372
373 static struct attribute *veth_cnx_default_attrs[] = {
374 GET_CNX_ATTR(outstanding_tx),
375 GET_CNX_ATTR(remote_lp),
376 GET_CNX_ATTR(num_events),
377 GET_CNX_ATTR(reset_timeout),
378 GET_CNX_ATTR(last_contact),
379 GET_CNX_ATTR(state),
380 GET_CNX_ATTR(src_inst),
381 GET_CNX_ATTR(dst_inst),
382 GET_CNX_ATTR(num_pending_acks),
383 GET_CNX_ATTR(num_ack_events),
384 GET_CNX_ATTR(ack_timeout),
385 NULL
386 };
387
388 static const struct sysfs_ops veth_cnx_sysfs_ops = {
389 .show = veth_cnx_attribute_show
390 };
391
392 static struct kobj_type veth_lpar_connection_ktype = {
393 .release = veth_release_connection,
394 .sysfs_ops = &veth_cnx_sysfs_ops,
395 .default_attrs = veth_cnx_default_attrs
396 };
397
398 struct veth_port_attribute {
399 struct attribute attr;
400 ssize_t (*show)(struct veth_port *, char *buf);
401 ssize_t (*store)(struct veth_port *, const char *buf);
402 };
403
404 static ssize_t veth_port_attribute_show(struct kobject *kobj,
405 struct attribute *attr, char *buf)
406 {
407 struct veth_port_attribute *port_attr;
408 struct veth_port *port;
409
410 port_attr = container_of(attr, struct veth_port_attribute, attr);
411 port = container_of(kobj, struct veth_port, kobject);
412
413 if (!port_attr->show)
414 return -EIO;
415
416 return port_attr->show(port, buf);
417 }
418
419 #define CUSTOM_PORT_ATTR(_name, _format, _expression) \
420 static ssize_t _name##_show(struct veth_port *port, char *buf) \
421 { \
422 return sprintf(buf, _format, _expression); \
423 } \
424 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
425
426 #define SIMPLE_PORT_ATTR(_name) \
427 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
428
429 SIMPLE_PORT_ATTR(promiscuous);
430 SIMPLE_PORT_ATTR(num_mcast);
431 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
432 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
433 CUSTOM_PORT_ATTR(mac_addr, "0x%llX\n", port->mac_addr);
434
435 #define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
436 static struct attribute *veth_port_default_attrs[] = {
437 GET_PORT_ATTR(mac_addr),
438 GET_PORT_ATTR(lpar_map),
439 GET_PORT_ATTR(stopped_map),
440 GET_PORT_ATTR(promiscuous),
441 GET_PORT_ATTR(num_mcast),
442 NULL
443 };
444
445 static const struct sysfs_ops veth_port_sysfs_ops = {
446 .show = veth_port_attribute_show
447 };
448
449 static struct kobj_type veth_port_ktype = {
450 .sysfs_ops = &veth_port_sysfs_ops,
451 .default_attrs = veth_port_default_attrs
452 };
453
454 /*
455 * LPAR connection code
456 */
457
458 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
459 {
460 schedule_delayed_work(&cnx->statemachine_wq, 0);
461 }
462
463 static void veth_take_cap(struct veth_lpar_connection *cnx,
464 struct veth_lpevent *event)
465 {
466 unsigned long flags;
467
468 spin_lock_irqsave(&cnx->lock, flags);
469 /* Receiving caps may mean the other end has just come up, so
470 * we need to reload the instance ID of the far end */
471 cnx->dst_inst =
472 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
473 HvLpEvent_Type_VirtualLan);
474
475 if (cnx->state & VETH_STATE_GOTCAPS) {
476 veth_error("Received a second capabilities from LPAR %d.\n",
477 cnx->remote_lp);
478 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
479 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
480 } else {
481 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
482 cnx->state |= VETH_STATE_GOTCAPS;
483 veth_kick_statemachine(cnx);
484 }
485 spin_unlock_irqrestore(&cnx->lock, flags);
486 }
487
488 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
489 struct veth_lpevent *event)
490 {
491 unsigned long flags;
492
493 spin_lock_irqsave(&cnx->lock, flags);
494 if (cnx->state & VETH_STATE_GOTCAPACK) {
495 veth_error("Received a second capabilities ack from LPAR %d.\n",
496 cnx->remote_lp);
497 } else {
498 memcpy(&cnx->cap_ack_event, event,
499 sizeof(cnx->cap_ack_event));
500 cnx->state |= VETH_STATE_GOTCAPACK;
501 veth_kick_statemachine(cnx);
502 }
503 spin_unlock_irqrestore(&cnx->lock, flags);
504 }
505
506 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
507 struct veth_lpevent *event)
508 {
509 unsigned long flags;
510
511 spin_lock_irqsave(&cnx->lock, flags);
512 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
513
514 /* Avoid kicking the statemachine once we're shutdown.
515 * It's unnecessary and it could break veth_stop_connection(). */
516
517 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
518 cnx->state |= VETH_STATE_RESET;
519 veth_kick_statemachine(cnx);
520 }
521 spin_unlock_irqrestore(&cnx->lock, flags);
522 }
523
524 static void veth_handle_ack(struct veth_lpevent *event)
525 {
526 HvLpIndex rlp = event->base_event.xTargetLp;
527 struct veth_lpar_connection *cnx = veth_cnx[rlp];
528
529 BUG_ON(! cnx);
530
531 switch (event->base_event.xSubtype) {
532 case VETH_EVENT_CAP:
533 veth_take_cap_ack(cnx, event);
534 break;
535 case VETH_EVENT_MONITOR:
536 veth_take_monitor_ack(cnx, event);
537 break;
538 default:
539 veth_error("Unknown ack type %d from LPAR %d.\n",
540 event->base_event.xSubtype, rlp);
541 };
542 }
543
544 static void veth_handle_int(struct veth_lpevent *event)
545 {
546 HvLpIndex rlp = event->base_event.xSourceLp;
547 struct veth_lpar_connection *cnx = veth_cnx[rlp];
548 unsigned long flags;
549 int i, acked = 0;
550
551 BUG_ON(! cnx);
552
553 switch (event->base_event.xSubtype) {
554 case VETH_EVENT_CAP:
555 veth_take_cap(cnx, event);
556 break;
557 case VETH_EVENT_MONITOR:
558 /* do nothing... this'll hang out here til we're dead,
559 * and the hypervisor will return it for us. */
560 break;
561 case VETH_EVENT_FRAMES_ACK:
562 spin_lock_irqsave(&cnx->lock, flags);
563
564 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
565 u16 msgnum = event->u.frames_ack_data.token[i];
566
567 if (msgnum < VETH_NUMBUFFERS) {
568 veth_recycle_msg(cnx, cnx->msgs + msgnum);
569 cnx->outstanding_tx--;
570 acked++;
571 }
572 }
573
574 if (acked > 0) {
575 cnx->last_contact = jiffies;
576 veth_wake_queues(cnx);
577 }
578
579 spin_unlock_irqrestore(&cnx->lock, flags);
580 break;
581 case VETH_EVENT_FRAMES:
582 veth_receive(cnx, event);
583 break;
584 default:
585 veth_error("Unknown interrupt type %d from LPAR %d.\n",
586 event->base_event.xSubtype, rlp);
587 };
588 }
589
590 static void veth_handle_event(struct HvLpEvent *event)
591 {
592 struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
593
594 if (hvlpevent_is_ack(event))
595 veth_handle_ack(veth_event);
596 else
597 veth_handle_int(veth_event);
598 }
599
600 static int veth_process_caps(struct veth_lpar_connection *cnx)
601 {
602 struct veth_cap_data *remote_caps = &cnx->remote_caps;
603 int num_acks_needed;
604
605 /* Convert timer to jiffies */
606 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
607
608 if ( (remote_caps->num_buffers == 0) ||
609 (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) ||
610 (remote_caps->ack_threshold == 0) ||
611 (cnx->ack_timeout == 0) ) {
612 veth_error("Received incompatible capabilities from LPAR %d.\n",
613 cnx->remote_lp);
614 return HvLpEvent_Rc_InvalidSubtypeData;
615 }
616
617 num_acks_needed = (remote_caps->num_buffers
618 / remote_caps->ack_threshold) + 1;
619
620 /* FIXME: locking on num_ack_events? */
621 if (cnx->num_ack_events < num_acks_needed) {
622 int num;
623
624 num = veth_allocate_events(cnx->remote_lp,
625 num_acks_needed-cnx->num_ack_events);
626 if (num > 0)
627 cnx->num_ack_events += num;
628
629 if (cnx->num_ack_events < num_acks_needed) {
630 veth_error("Couldn't allocate enough ack events "
631 "for LPAR %d.\n", cnx->remote_lp);
632
633 return HvLpEvent_Rc_BufferNotAvailable;
634 }
635 }
636
637
638 return HvLpEvent_Rc_Good;
639 }
640
641 /* FIXME: The gotos here are a bit dubious */
642 static void veth_statemachine(struct work_struct *work)
643 {
644 struct veth_lpar_connection *cnx =
645 container_of(work, struct veth_lpar_connection,
646 statemachine_wq.work);
647 int rlp = cnx->remote_lp;
648 int rc;
649
650 spin_lock_irq(&cnx->lock);
651
652 restart:
653 if (cnx->state & VETH_STATE_RESET) {
654 if (cnx->state & VETH_STATE_OPEN)
655 HvCallEvent_closeLpEventPath(cnx->remote_lp,
656 HvLpEvent_Type_VirtualLan);
657
658 /*
659 * Reset ack data. This prevents the ack_timer actually
660 * doing anything, even if it runs one more time when
661 * we drop the lock below.
662 */
663 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
664 cnx->num_pending_acks = 0;
665
666 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
667 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
668 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
669 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
670
671 /* Clean up any leftover messages */
672 if (cnx->msgs) {
673 int i;
674 for (i = 0; i < VETH_NUMBUFFERS; ++i)
675 veth_recycle_msg(cnx, cnx->msgs + i);
676 }
677
678 cnx->outstanding_tx = 0;
679 veth_wake_queues(cnx);
680
681 /* Drop the lock so we can do stuff that might sleep or
682 * take other locks. */
683 spin_unlock_irq(&cnx->lock);
684
685 del_timer_sync(&cnx->ack_timer);
686 del_timer_sync(&cnx->reset_timer);
687
688 spin_lock_irq(&cnx->lock);
689
690 if (cnx->state & VETH_STATE_RESET)
691 goto restart;
692
693 /* Hack, wait for the other end to reset itself. */
694 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
695 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
696 goto out;
697 }
698 }
699
700 if (cnx->state & VETH_STATE_SHUTDOWN)
701 /* It's all over, do nothing */
702 goto out;
703
704 if ( !(cnx->state & VETH_STATE_OPEN) ) {
705 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
706 goto cant_cope;
707
708 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
709 cnx->src_inst =
710 HvCallEvent_getSourceLpInstanceId(rlp,
711 HvLpEvent_Type_VirtualLan);
712 cnx->dst_inst =
713 HvCallEvent_getTargetLpInstanceId(rlp,
714 HvLpEvent_Type_VirtualLan);
715 cnx->state |= VETH_STATE_OPEN;
716 }
717
718 if ( (cnx->state & VETH_STATE_OPEN) &&
719 !(cnx->state & VETH_STATE_SENTMON) ) {
720 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
721 HvLpEvent_AckInd_DoAck,
722 HvLpEvent_AckType_DeferredAck,
723 0, 0, 0, 0, 0, 0);
724
725 if (rc == HvLpEvent_Rc_Good) {
726 cnx->state |= VETH_STATE_SENTMON;
727 } else {
728 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
729 (rc != HvLpEvent_Rc_PathClosed) )
730 veth_error("Error sending monitor to LPAR %d, "
731 "rc = %d\n", rlp, rc);
732
733 /* Oh well, hope we get a cap from the other
734 * end and do better when that kicks us */
735 goto out;
736 }
737 }
738
739 if ( (cnx->state & VETH_STATE_OPEN) &&
740 !(cnx->state & VETH_STATE_SENTCAPS)) {
741 u64 *rawcap = (u64 *)&cnx->local_caps;
742
743 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
744 HvLpEvent_AckInd_DoAck,
745 HvLpEvent_AckType_ImmediateAck,
746 0, rawcap[0], rawcap[1], rawcap[2],
747 rawcap[3], rawcap[4]);
748
749 if (rc == HvLpEvent_Rc_Good) {
750 cnx->state |= VETH_STATE_SENTCAPS;
751 } else {
752 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
753 (rc != HvLpEvent_Rc_PathClosed) )
754 veth_error("Error sending caps to LPAR %d, "
755 "rc = %d\n", rlp, rc);
756
757 /* Oh well, hope we get a cap from the other
758 * end and do better when that kicks us */
759 goto out;
760 }
761 }
762
763 if ((cnx->state & VETH_STATE_GOTCAPS) &&
764 !(cnx->state & VETH_STATE_SENTCAPACK)) {
765 struct veth_cap_data *remote_caps = &cnx->remote_caps;
766
767 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
768 sizeof(*remote_caps));
769
770 spin_unlock_irq(&cnx->lock);
771 rc = veth_process_caps(cnx);
772 spin_lock_irq(&cnx->lock);
773
774 /* We dropped the lock, so recheck for anything which
775 * might mess us up */
776 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
777 goto restart;
778
779 cnx->cap_event.base_event.xRc = rc;
780 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
781 if (rc == HvLpEvent_Rc_Good)
782 cnx->state |= VETH_STATE_SENTCAPACK;
783 else
784 goto cant_cope;
785 }
786
787 if ((cnx->state & VETH_STATE_GOTCAPACK) &&
788 (cnx->state & VETH_STATE_GOTCAPS) &&
789 !(cnx->state & VETH_STATE_READY)) {
790 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
791 /* Start the ACK timer */
792 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
793 add_timer(&cnx->ack_timer);
794 cnx->state |= VETH_STATE_READY;
795 } else {
796 veth_error("Caps rejected by LPAR %d, rc = %d\n",
797 rlp, cnx->cap_ack_event.base_event.xRc);
798 goto cant_cope;
799 }
800 }
801
802 out:
803 spin_unlock_irq(&cnx->lock);
804 return;
805
806 cant_cope:
807 /* FIXME: we get here if something happens we really can't
808 * cope with. The link will never work once we get here, and
809 * all we can do is not lock the rest of the system up */
810 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
811 " (state = 0x%04lx)\n", rlp, cnx->state);
812 cnx->state |= VETH_STATE_SHUTDOWN;
813 spin_unlock_irq(&cnx->lock);
814 }
815
816 static int veth_init_connection(u8 rlp)
817 {
818 struct veth_lpar_connection *cnx;
819 struct veth_msg *msgs;
820 int i;
821
822 if ( (rlp == this_lp) ||
823 ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
824 return 0;
825
826 cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
827 if (! cnx)
828 return -ENOMEM;
829
830 cnx->remote_lp = rlp;
831 spin_lock_init(&cnx->lock);
832 INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
833
834 init_timer(&cnx->ack_timer);
835 cnx->ack_timer.function = veth_timed_ack;
836 cnx->ack_timer.data = (unsigned long) cnx;
837
838 init_timer(&cnx->reset_timer);
839 cnx->reset_timer.function = veth_timed_reset;
840 cnx->reset_timer.data = (unsigned long) cnx;
841 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
842
843 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
844
845 veth_cnx[rlp] = cnx;
846
847 /* This gets us 1 reference, which is held on behalf of the driver
848 * infrastructure. It's released at module unload. */
849 kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
850
851 msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
852 if (! msgs) {
853 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
854 return -ENOMEM;
855 }
856
857 cnx->msgs = msgs;
858
859 for (i = 0; i < VETH_NUMBUFFERS; i++) {
860 msgs[i].token = i;
861 veth_stack_push(cnx, msgs + i);
862 }
863
864 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
865
866 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
867 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
868 return -ENOMEM;
869 }
870
871 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
872 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
873 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
874
875 return 0;
876 }
877
878 static void veth_stop_connection(struct veth_lpar_connection *cnx)
879 {
880 if (!cnx)
881 return;
882
883 spin_lock_irq(&cnx->lock);
884 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
885 veth_kick_statemachine(cnx);
886 spin_unlock_irq(&cnx->lock);
887
888 /* ensure the statemachine runs now and waits for its completion */
889 flush_delayed_work_sync(&cnx->statemachine_wq);
890 }
891
892 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
893 {
894 if (!cnx)
895 return;
896
897 if (cnx->num_events > 0)
898 mf_deallocate_lp_events(cnx->remote_lp,
899 HvLpEvent_Type_VirtualLan,
900 cnx->num_events,
901 NULL, NULL);
902 if (cnx->num_ack_events > 0)
903 mf_deallocate_lp_events(cnx->remote_lp,
904 HvLpEvent_Type_VirtualLan,
905 cnx->num_ack_events,
906 NULL, NULL);
907
908 kfree(cnx->msgs);
909 veth_cnx[cnx->remote_lp] = NULL;
910 kfree(cnx);
911 }
912
913 static void veth_release_connection(struct kobject *kobj)
914 {
915 struct veth_lpar_connection *cnx;
916 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
917 veth_stop_connection(cnx);
918 veth_destroy_connection(cnx);
919 }
920
921 /*
922 * net_device code
923 */
924
925 static int veth_open(struct net_device *dev)
926 {
927 netif_start_queue(dev);
928 return 0;
929 }
930
931 static int veth_close(struct net_device *dev)
932 {
933 netif_stop_queue(dev);
934 return 0;
935 }
936
937 static int veth_change_mtu(struct net_device *dev, int new_mtu)
938 {
939 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
940 return -EINVAL;
941 dev->mtu = new_mtu;
942 return 0;
943 }
944
945 static void veth_set_multicast_list(struct net_device *dev)
946 {
947 struct veth_port *port = netdev_priv(dev);
948 unsigned long flags;
949
950 write_lock_irqsave(&port->mcast_gate, flags);
951
952 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
953 (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
954 port->promiscuous = 1;
955 } else {
956 struct netdev_hw_addr *ha;
957
958 port->promiscuous = 0;
959
960 /* Update table */
961 port->num_mcast = 0;
962
963 netdev_for_each_mc_addr(ha, dev) {
964 u8 *addr = ha->addr;
965 u64 xaddr = 0;
966
967 if (addr[0] & 0x01) {/* multicast address? */
968 memcpy(&xaddr, addr, ETH_ALEN);
969 port->mcast_addr[port->num_mcast] = xaddr;
970 port->num_mcast++;
971 }
972 }
973 }
974
975 write_unlock_irqrestore(&port->mcast_gate, flags);
976 }
977
978 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
979 {
980 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
981 info->driver[sizeof(info->driver) - 1] = '\0';
982 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
983 info->version[sizeof(info->version) - 1] = '\0';
984 }
985
986 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
987 {
988 ecmd->supported = (SUPPORTED_1000baseT_Full
989 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
990 ecmd->advertising = (SUPPORTED_1000baseT_Full
991 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
992 ecmd->port = PORT_FIBRE;
993 ecmd->transceiver = XCVR_INTERNAL;
994 ecmd->phy_address = 0;
995 ecmd->speed = SPEED_1000;
996 ecmd->duplex = DUPLEX_FULL;
997 ecmd->autoneg = AUTONEG_ENABLE;
998 ecmd->maxtxpkt = 120;
999 ecmd->maxrxpkt = 120;
1000 return 0;
1001 }
1002
1003 static const struct ethtool_ops ops = {
1004 .get_drvinfo = veth_get_drvinfo,
1005 .get_settings = veth_get_settings,
1006 .get_link = ethtool_op_get_link,
1007 };
1008
1009 static const struct net_device_ops veth_netdev_ops = {
1010 .ndo_open = veth_open,
1011 .ndo_stop = veth_close,
1012 .ndo_start_xmit = veth_start_xmit,
1013 .ndo_change_mtu = veth_change_mtu,
1014 .ndo_set_multicast_list = veth_set_multicast_list,
1015 .ndo_set_mac_address = NULL,
1016 .ndo_validate_addr = eth_validate_addr,
1017 };
1018
1019 static struct net_device *veth_probe_one(int vlan,
1020 struct vio_dev *vio_dev)
1021 {
1022 struct net_device *dev;
1023 struct veth_port *port;
1024 struct device *vdev = &vio_dev->dev;
1025 int i, rc;
1026 const unsigned char *mac_addr;
1027
1028 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1029 if (mac_addr == NULL)
1030 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1031 if (mac_addr == NULL) {
1032 veth_error("Unable to fetch MAC address from device tree.\n");
1033 return NULL;
1034 }
1035
1036 dev = alloc_etherdev(sizeof (struct veth_port));
1037 if (! dev) {
1038 veth_error("Unable to allocate net_device structure!\n");
1039 return NULL;
1040 }
1041
1042 port = netdev_priv(dev);
1043
1044 spin_lock_init(&port->queue_lock);
1045 rwlock_init(&port->mcast_gate);
1046 port->stopped_map = 0;
1047
1048 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1049 HvLpVirtualLanIndexMap map;
1050
1051 if (i == this_lp)
1052 continue;
1053 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1054 if (map & (0x8000 >> vlan))
1055 port->lpar_map |= (1 << i);
1056 }
1057 port->dev = vdev;
1058
1059 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
1060
1061 dev->mtu = VETH_MAX_MTU;
1062
1063 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
1064
1065 dev->netdev_ops = &veth_netdev_ops;
1066 SET_ETHTOOL_OPS(dev, &ops);
1067
1068 SET_NETDEV_DEV(dev, vdev);
1069
1070 rc = register_netdev(dev);
1071 if (rc != 0) {
1072 veth_error("Failed registering net device for vlan%d.\n", vlan);
1073 free_netdev(dev);
1074 return NULL;
1075 }
1076
1077 kobject_init(&port->kobject, &veth_port_ktype);
1078 if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
1079 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1080
1081 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1082 dev->name, vlan, port->lpar_map);
1083
1084 return dev;
1085 }
1086
1087 /*
1088 * Tx path
1089 */
1090
1091 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1092 struct net_device *dev)
1093 {
1094 struct veth_lpar_connection *cnx = veth_cnx[rlp];
1095 struct veth_port *port = netdev_priv(dev);
1096 HvLpEvent_Rc rc;
1097 struct veth_msg *msg = NULL;
1098 unsigned long flags;
1099
1100 if (! cnx)
1101 return 0;
1102
1103 spin_lock_irqsave(&cnx->lock, flags);
1104
1105 if (! (cnx->state & VETH_STATE_READY))
1106 goto no_error;
1107
1108 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1109 goto drop;
1110
1111 msg = veth_stack_pop(cnx);
1112 if (! msg)
1113 goto drop;
1114
1115 msg->in_use = 1;
1116 msg->skb = skb_get(skb);
1117
1118 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1119 skb->len, DMA_TO_DEVICE);
1120
1121 if (dma_mapping_error(port->dev, msg->data.addr[0]))
1122 goto recycle_and_drop;
1123
1124 msg->dev = port->dev;
1125 msg->data.len[0] = skb->len;
1126 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1127
1128 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1129
1130 if (rc != HvLpEvent_Rc_Good)
1131 goto recycle_and_drop;
1132
1133 /* If the timer's not already running, start it now. */
1134 if (0 == cnx->outstanding_tx)
1135 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1136
1137 cnx->last_contact = jiffies;
1138 cnx->outstanding_tx++;
1139
1140 if (veth_stack_is_empty(cnx))
1141 veth_stop_queues(cnx);
1142
1143 no_error:
1144 spin_unlock_irqrestore(&cnx->lock, flags);
1145 return 0;
1146
1147 recycle_and_drop:
1148 veth_recycle_msg(cnx, msg);
1149 drop:
1150 spin_unlock_irqrestore(&cnx->lock, flags);
1151 return 1;
1152 }
1153
1154 static void veth_transmit_to_many(struct sk_buff *skb,
1155 HvLpIndexMap lpmask,
1156 struct net_device *dev)
1157 {
1158 int i, success, error;
1159
1160 success = error = 0;
1161
1162 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1163 if ((lpmask & (1 << i)) == 0)
1164 continue;
1165
1166 if (veth_transmit_to_one(skb, i, dev))
1167 error = 1;
1168 else
1169 success = 1;
1170 }
1171
1172 if (error)
1173 dev->stats.tx_errors++;
1174
1175 if (success) {
1176 dev->stats.tx_packets++;
1177 dev->stats.tx_bytes += skb->len;
1178 }
1179 }
1180
1181 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1182 {
1183 unsigned char *frame = skb->data;
1184 struct veth_port *port = netdev_priv(dev);
1185 HvLpIndexMap lpmask;
1186
1187 if (! (frame[0] & 0x01)) {
1188 /* unicast packet */
1189 HvLpIndex rlp = frame[5];
1190
1191 if ( ! ((1 << rlp) & port->lpar_map) ) {
1192 dev_kfree_skb(skb);
1193 return NETDEV_TX_OK;
1194 }
1195
1196 lpmask = 1 << rlp;
1197 } else {
1198 lpmask = port->lpar_map;
1199 }
1200
1201 veth_transmit_to_many(skb, lpmask, dev);
1202
1203 dev_kfree_skb(skb);
1204
1205 return NETDEV_TX_OK;
1206 }
1207
1208 /* You must hold the connection's lock when you call this function. */
1209 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1210 struct veth_msg *msg)
1211 {
1212 u32 dma_address, dma_length;
1213
1214 if (msg->in_use) {
1215 msg->in_use = 0;
1216 dma_address = msg->data.addr[0];
1217 dma_length = msg->data.len[0];
1218
1219 if (!dma_mapping_error(msg->dev, dma_address))
1220 dma_unmap_single(msg->dev, dma_address, dma_length,
1221 DMA_TO_DEVICE);
1222
1223 if (msg->skb) {
1224 dev_kfree_skb_any(msg->skb);
1225 msg->skb = NULL;
1226 }
1227
1228 memset(&msg->data, 0, sizeof(msg->data));
1229 veth_stack_push(cnx, msg);
1230 } else if (cnx->state & VETH_STATE_OPEN) {
1231 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1232 cnx->remote_lp, msg->token);
1233 }
1234 }
1235
1236 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1237 {
1238 int i;
1239
1240 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1241 struct net_device *dev = veth_dev[i];
1242 struct veth_port *port;
1243 unsigned long flags;
1244
1245 if (! dev)
1246 continue;
1247
1248 port = netdev_priv(dev);
1249
1250 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1251 continue;
1252
1253 spin_lock_irqsave(&port->queue_lock, flags);
1254
1255 port->stopped_map &= ~(1 << cnx->remote_lp);
1256
1257 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1258 veth_debug("cnx %d: woke queue for %s.\n",
1259 cnx->remote_lp, dev->name);
1260 netif_wake_queue(dev);
1261 }
1262 spin_unlock_irqrestore(&port->queue_lock, flags);
1263 }
1264 }
1265
1266 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1267 {
1268 int i;
1269
1270 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1271 struct net_device *dev = veth_dev[i];
1272 struct veth_port *port;
1273
1274 if (! dev)
1275 continue;
1276
1277 port = netdev_priv(dev);
1278
1279 /* If this cnx is not on the vlan for this port, continue */
1280 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1281 continue;
1282
1283 spin_lock(&port->queue_lock);
1284
1285 netif_stop_queue(dev);
1286 port->stopped_map |= (1 << cnx->remote_lp);
1287
1288 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1289 cnx->remote_lp, dev->name, port->stopped_map);
1290
1291 spin_unlock(&port->queue_lock);
1292 }
1293 }
1294
1295 static void veth_timed_reset(unsigned long ptr)
1296 {
1297 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1298 unsigned long trigger_time, flags;
1299
1300 /* FIXME is it possible this fires after veth_stop_connection()?
1301 * That would reschedule the statemachine for 5 seconds and probably
1302 * execute it after the module's been unloaded. Hmm. */
1303
1304 spin_lock_irqsave(&cnx->lock, flags);
1305
1306 if (cnx->outstanding_tx > 0) {
1307 trigger_time = cnx->last_contact + cnx->reset_timeout;
1308
1309 if (trigger_time < jiffies) {
1310 cnx->state |= VETH_STATE_RESET;
1311 veth_kick_statemachine(cnx);
1312 veth_error("%d packets not acked by LPAR %d within %d "
1313 "seconds, resetting.\n",
1314 cnx->outstanding_tx, cnx->remote_lp,
1315 cnx->reset_timeout / HZ);
1316 } else {
1317 /* Reschedule the timer */
1318 trigger_time = jiffies + cnx->reset_timeout;
1319 mod_timer(&cnx->reset_timer, trigger_time);
1320 }
1321 }
1322
1323 spin_unlock_irqrestore(&cnx->lock, flags);
1324 }
1325
1326 /*
1327 * Rx path
1328 */
1329
1330 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1331 {
1332 int wanted = 0;
1333 int i;
1334 unsigned long flags;
1335
1336 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1337 return 1;
1338
1339 read_lock_irqsave(&port->mcast_gate, flags);
1340
1341 if (port->promiscuous) {
1342 wanted = 1;
1343 goto out;
1344 }
1345
1346 for (i = 0; i < port->num_mcast; ++i) {
1347 if (port->mcast_addr[i] == mac_addr) {
1348 wanted = 1;
1349 break;
1350 }
1351 }
1352
1353 out:
1354 read_unlock_irqrestore(&port->mcast_gate, flags);
1355
1356 return wanted;
1357 }
1358
1359 struct dma_chunk {
1360 u64 addr;
1361 u64 size;
1362 };
1363
1364 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1365
1366 static inline void veth_build_dma_list(struct dma_chunk *list,
1367 unsigned char *p, unsigned long length)
1368 {
1369 unsigned long done;
1370 int i = 1;
1371
1372 /* FIXME: skbs are contiguous in real addresses. Do we
1373 * really need to break it into PAGE_SIZE chunks, or can we do
1374 * it just at the granularity of iSeries real->absolute
1375 * mapping? Indeed, given the way the allocator works, can we
1376 * count on them being absolutely contiguous? */
1377 list[0].addr = iseries_hv_addr(p);
1378 list[0].size = min(length,
1379 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1380
1381 done = list[0].size;
1382 while (done < length) {
1383 list[i].addr = iseries_hv_addr(p + done);
1384 list[i].size = min(length-done, PAGE_SIZE);
1385 done += list[i].size;
1386 i++;
1387 }
1388 }
1389
1390 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1391 {
1392 HvLpEvent_Rc rc;
1393
1394 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1395 0, &cnx->pending_acks);
1396
1397 if (rc != HvLpEvent_Rc_Good)
1398 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1399 cnx->remote_lp, (int)rc);
1400
1401 cnx->num_pending_acks = 0;
1402 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1403 }
1404
1405 static void veth_receive(struct veth_lpar_connection *cnx,
1406 struct veth_lpevent *event)
1407 {
1408 struct veth_frames_data *senddata = &event->u.frames_data;
1409 int startchunk = 0;
1410 int nchunks;
1411 unsigned long flags;
1412 HvLpDma_Rc rc;
1413
1414 do {
1415 u16 length = 0;
1416 struct sk_buff *skb;
1417 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1418 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1419 u64 dest;
1420 HvLpVirtualLanIndex vlan;
1421 struct net_device *dev;
1422 struct veth_port *port;
1423
1424 /* FIXME: do we need this? */
1425 memset(local_list, 0, sizeof(local_list));
1426 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1427
1428 /* a 0 address marks the end of the valid entries */
1429 if (senddata->addr[startchunk] == 0)
1430 break;
1431
1432 /* make sure that we have at least 1 EOF entry in the
1433 * remaining entries */
1434 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1435 veth_error("Missing EOF fragment in event "
1436 "eofmask = 0x%x startchunk = %d\n",
1437 (unsigned)senddata->eofmask,
1438 startchunk);
1439 break;
1440 }
1441
1442 /* build list of chunks in this frame */
1443 nchunks = 0;
1444 do {
1445 remote_list[nchunks].addr =
1446 (u64) senddata->addr[startchunk+nchunks] << 32;
1447 remote_list[nchunks].size =
1448 senddata->len[startchunk+nchunks];
1449 length += remote_list[nchunks].size;
1450 } while (! (senddata->eofmask &
1451 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1452
1453 /* length == total length of all chunks */
1454 /* nchunks == # of chunks in this frame */
1455
1456 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1457 veth_error("Received oversize frame from LPAR %d "
1458 "(length = %d)\n",
1459 cnx->remote_lp, length);
1460 continue;
1461 }
1462
1463 skb = alloc_skb(length, GFP_ATOMIC);
1464 if (!skb)
1465 continue;
1466
1467 veth_build_dma_list(local_list, skb->data, length);
1468
1469 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1470 event->base_event.xSourceLp,
1471 HvLpDma_Direction_RemoteToLocal,
1472 cnx->src_inst,
1473 cnx->dst_inst,
1474 HvLpDma_AddressType_RealAddress,
1475 HvLpDma_AddressType_TceIndex,
1476 iseries_hv_addr(&local_list),
1477 iseries_hv_addr(&remote_list),
1478 length);
1479 if (rc != HvLpDma_Rc_Good) {
1480 dev_kfree_skb_irq(skb);
1481 continue;
1482 }
1483
1484 vlan = skb->data[9];
1485 dev = veth_dev[vlan];
1486 if (! dev) {
1487 /*
1488 * Some earlier versions of the driver sent
1489 * broadcasts down all connections, even to lpars
1490 * that weren't on the relevant vlan. So ignore
1491 * packets belonging to a vlan we're not on.
1492 * We can also be here if we receive packets while
1493 * the driver is going down, because then dev is NULL.
1494 */
1495 dev_kfree_skb_irq(skb);
1496 continue;
1497 }
1498
1499 port = netdev_priv(dev);
1500 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1501
1502 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1503 dev_kfree_skb_irq(skb);
1504 continue;
1505 }
1506 if (! veth_frame_wanted(port, dest)) {
1507 dev_kfree_skb_irq(skb);
1508 continue;
1509 }
1510
1511 skb_put(skb, length);
1512 skb->protocol = eth_type_trans(skb, dev);
1513 skb_checksum_none_assert(skb);
1514 netif_rx(skb); /* send it up */
1515 dev->stats.rx_packets++;
1516 dev->stats.rx_bytes += length;
1517 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1518
1519 /* Ack it */
1520 spin_lock_irqsave(&cnx->lock, flags);
1521 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1522
1523 cnx->pending_acks[cnx->num_pending_acks++] =
1524 event->base_event.xCorrelationToken;
1525
1526 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1527 (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1528 veth_flush_acks(cnx);
1529
1530 spin_unlock_irqrestore(&cnx->lock, flags);
1531 }
1532
1533 static void veth_timed_ack(unsigned long ptr)
1534 {
1535 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1536 unsigned long flags;
1537
1538 /* Ack all the events */
1539 spin_lock_irqsave(&cnx->lock, flags);
1540 if (cnx->num_pending_acks > 0)
1541 veth_flush_acks(cnx);
1542
1543 /* Reschedule the timer */
1544 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1545 add_timer(&cnx->ack_timer);
1546 spin_unlock_irqrestore(&cnx->lock, flags);
1547 }
1548
1549 static int veth_remove(struct vio_dev *vdev)
1550 {
1551 struct veth_lpar_connection *cnx;
1552 struct net_device *dev;
1553 struct veth_port *port;
1554 int i;
1555
1556 dev = veth_dev[vdev->unit_address];
1557
1558 if (! dev)
1559 return 0;
1560
1561 port = netdev_priv(dev);
1562
1563 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1564 cnx = veth_cnx[i];
1565
1566 if (cnx && (port->lpar_map & (1 << i))) {
1567 /* Drop our reference to connections on our VLAN */
1568 kobject_put(&cnx->kobject);
1569 }
1570 }
1571
1572 veth_dev[vdev->unit_address] = NULL;
1573 kobject_del(&port->kobject);
1574 kobject_put(&port->kobject);
1575 unregister_netdev(dev);
1576 free_netdev(dev);
1577
1578 return 0;
1579 }
1580
1581 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1582 {
1583 int i = vdev->unit_address;
1584 struct net_device *dev;
1585 struct veth_port *port;
1586
1587 dev = veth_probe_one(i, vdev);
1588 if (dev == NULL) {
1589 veth_remove(vdev);
1590 return 1;
1591 }
1592 veth_dev[i] = dev;
1593
1594 port = netdev_priv(dev);
1595
1596 /* Start the state machine on each connection on this vlan. If we're
1597 * the first dev to do so this will commence link negotiation */
1598 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1599 struct veth_lpar_connection *cnx;
1600
1601 if (! (port->lpar_map & (1 << i)))
1602 continue;
1603
1604 cnx = veth_cnx[i];
1605 if (!cnx)
1606 continue;
1607
1608 kobject_get(&cnx->kobject);
1609 veth_kick_statemachine(cnx);
1610 }
1611
1612 return 0;
1613 }
1614
1615 /**
1616 * veth_device_table: Used by vio.c to match devices that we
1617 * support.
1618 */
1619 static struct vio_device_id veth_device_table[] __devinitdata = {
1620 { "network", "IBM,iSeries-l-lan" },
1621 { "", "" }
1622 };
1623 MODULE_DEVICE_TABLE(vio, veth_device_table);
1624
1625 static struct vio_driver veth_driver = {
1626 .id_table = veth_device_table,
1627 .probe = veth_probe,
1628 .remove = veth_remove,
1629 .driver = {
1630 .name = DRV_NAME,
1631 .owner = THIS_MODULE,
1632 }
1633 };
1634
1635 /*
1636 * Module initialization/cleanup
1637 */
1638
1639 static void __exit veth_module_cleanup(void)
1640 {
1641 int i;
1642 struct veth_lpar_connection *cnx;
1643
1644 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1645 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1646
1647 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1648 cnx = veth_cnx[i];
1649
1650 if (!cnx)
1651 continue;
1652
1653 /* Cancel work queued from Hypervisor callbacks */
1654 cancel_delayed_work_sync(&cnx->statemachine_wq);
1655 /* Remove the connection from sysfs */
1656 kobject_del(&cnx->kobject);
1657 /* Drop the driver's reference to the connection */
1658 kobject_put(&cnx->kobject);
1659 }
1660
1661 /* Unregister the driver, which will close all the netdevs and stop
1662 * the connections when they're no longer referenced. */
1663 vio_unregister_driver(&veth_driver);
1664 }
1665 module_exit(veth_module_cleanup);
1666
1667 static int __init veth_module_init(void)
1668 {
1669 int i;
1670 int rc;
1671
1672 if (!firmware_has_feature(FW_FEATURE_ISERIES))
1673 return -ENODEV;
1674
1675 this_lp = HvLpConfig_getLpIndex_outline();
1676
1677 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1678 rc = veth_init_connection(i);
1679 if (rc != 0)
1680 goto error;
1681 }
1682
1683 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1684 &veth_handle_event);
1685
1686 rc = vio_register_driver(&veth_driver);
1687 if (rc != 0)
1688 goto error;
1689
1690 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1691 struct kobject *kobj;
1692
1693 if (!veth_cnx[i])
1694 continue;
1695
1696 kobj = &veth_cnx[i]->kobject;
1697 /* If the add failes, complain but otherwise continue */
1698 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
1699 "cnx%.2d", veth_cnx[i]->remote_lp))
1700 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1701 }
1702
1703 return 0;
1704
1705 error:
1706 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1707 veth_destroy_connection(veth_cnx[i]);
1708 }
1709
1710 return rc;
1711 }
1712 module_init(veth_module_init);
linux-next